goalforge-redecomposelisted

# SDD Redecompose Given a learning and a (possibly partially-executed) feature, re-runs WP decomposition and reconciles the result against the feature's existing WPs — some of which may already be verified. This is the keystone agile loop-back: a new learning may invalidate some WPs, and the skill determines exactly which ones change, routes them back to harden, and preserves everything that is already verified. ## Preconditions - A feature directory exists under `plans/` with `wp-*/overview.md` files. - A proposed re-decomposition is available as a JSON array `[{"slug":"...","goal_outcome":"..."}, ...]`. - The caller supplies an explicit **trigger param** (`--learning "<text>"` or `trigger_reason`) identifying what learning prompted the re-decomposition. This param is intentionally open — not hard-wired to human-only invocation — so a downstream automated edge (e.g., a pipeline detecting a newly-verified upstream WP) can invoke goalforge-redecompose as well. Always capture the trigger_reason for the ledger. ## Inputs | Param | Description | |-------|-------------| | `<feature-dir>` | Absolute or plans-root-relative path to the feature directory | | `<proposed-json>` | Path to JSON file: `[{"slug":"...","goal_outcome":"..."}, ...]` | | `--learning "<text>"` | The trigger reason / learning that prompted re-decomposition | ## Step 1 — Call the pure diff Run the deterministic reconcile-diff and capture its output as typed data: ```bash bash "${CLAUDE_PLUGIN_ROOT}